Yet More Fun With Radiosity.

A while back, I took people to task for heaping praise on some of
Gilles' rather less impressive work on radiosity, and pointed out that
any of us could have made those same radiosity tests.

Well, while I *am* the sort of guy who'll make comments like that
without any kind of substantial proof, I felt that some was needed, and
I recently found time to make some.

The image that follows is a rough (hand-modeled! bleh...) copy of the
front third or so of the image at
http://www.3dluvr.com/marcosss/sky6.jpg .

Notes:

1. This was done with MegaPOV radiosity, but I don't see why some
tweaking couldn't produce good results with the standard distribution.
Someone else can spend time on it, however. :P

2. While experimenting with different settings, I found that the default
radiosity DOESN'T CAST SHADOWS. Madness, I tell you, madness! By far the
most important change to the standard radiosity in this scene is setting
an absurdly low error_bound (0.3, I think) in the global block; it made
all the difference. Note that the shadows will be really patchy unless
you also use an absurdly low pretrace_end.

3. Amazingly, recursion was limited to 1. Higher levels produced
undesirable effects...

4. The raytracer that other guy uses seems to produce a LOT of color
bleed; even with an unnaturally vivid sky, I wasn't able to reproduce
his depth of saturation. Makes me wonder if negative gray_threshold
would be a useful feature (it's currently limited to positive values).

5. Yes, the grain comes from post-processing. Sue me.

6. Yes, I'll post the code in the text scene files group, as soon as I
get around to it.

7. Render time was 5 minutes, 35 seconds on a Power Mac 7100/66AV (with
G3/266 upgrade).

-Xplo

Post a reply to this message

Attachments:
Download 'discs-radiosity.jpg' (29 KB)

Preview of image 'discs-radiosity.jpg'